Portfolio movement, identified while the pricing window remains open
An annual review establishes that a portfolio has deteriorated, but reports it after the period responsible has closed. Prizmetrics analyses the claims and premium extracts an insurer already produces and identifies the movement while pricing and terms can still be adjusted.
Last updated 12 September 2026
Attribution of movement
A loss ratio that has moved by four points is an observation rather than a decision. The decision requires the group, network, benefit and age band responsible, and the proportion of the total each accounts for.
Every measure is therefore broken down consistently: by group and policy cohort, by network and provider, by benefit family and claim type, by nationality, and by age band and gender. Each contribution is reported as an absolute value alongside its share of the total, because a segment that has moved thirty per cent on a small base and one that has moved three per cent across half the portfolio represent different problems, and a percentage alone does not distinguish them.
Measure definitions
Disagreements about a loss ratio are usually disagreements about its denominator. The definitions are fixed once and carried with every figure derived from them:
- Earned premium is annual premium earned pro rata by day to the report date.
- Exposure is earned member-years to the report date.
- Burn cost is incurred claims per life-year, measured on that exposure.
- Estimated IBNR is calculated against a stated lag, which is reported with the figure.
The Excel workbook and the printable report state the report date, the assumption behind estimated IBNR and its day window, the target loss ratio, the medical inflation rate and the large-loss threshold on the face of the file. An analyst receiving the file three weeks later can establish the basis of every figure without further enquiry.
The report date
Changing the report date does not filter records. It re-earns premium to the new date, restates exposure, and recomputes every measure derived from either. A comparison between two dates is therefore a comparison of two complete views of the portfolio rather than of two subsets of one.
One consequence should be stated explicitly, as it is unexpected on first use: historical figures change when the report date changes. This reflects the calculation being applied correctly rather than instability in the underlying data.
Input requirements
The premium register and claims extract produced by the existing policy administration or claims system. No reformatting is required and no template has to be populated. Columns are matched against a defined vocabulary, any that cannot be resolved with confidence are presented for confirmation before the file is loaded, and the resulting mapping is retained so that subsequent refreshes require only the file. Extract requirements.
Identifying columns are deleted from each record before mapping and before any other code reads the row, rather than filtered out subsequently. The identifiers that remain, which join a claim to a premium record, are hashed, and only the hash is stored. Data handling.
Evaluation
A demonstration is conducted on the insurer's own extract rather than on sample data. Prizmetrics will confirm what the analysis would require, including any fields the extract does not currently contain.